Offline Transcription on Mac: Transcribe With No Internet

You’re on a flight with a recorded interview you need as text before you land. Or you’re in a basement conference room where the wifi drops every few minutes. Or you work somewhere — a clinic, a law office, a secure facility — where uploading audio to a website is simply not an option. In every one of those situations the same thing is true: you have audio, you need text, and the internet is either missing or off-limits. Most transcription tools fall over at exactly that point, because the transcription doesn’t actually happen on your machine. It happens on a server you can’t reach.

Offline transcription on a Mac means the whole job — turning speech into text — runs locally on your own hardware, with no network connection required and no audio sent anywhere. This post is about how that works on a modern Mac, why it’s now genuinely as good as the cloud version, and how to set it up so “transcribe this” stops depending on whether you have a signal.

Why most transcription tools need the internet

The reason so many transcription apps grind to a halt without wifi is architectural. Otter, Fireflies, and most browser-based transcribers are thin clients. The app on your screen records or accepts the audio, then ships it to the company’s backend, where the actual model runs. The transcript comes back over the same connection. No connection, no transcript. The “app” is really a remote control for a server.

That design made sense when on-device speech models were too heavy to run on a laptop. It doesn’t anymore, but the architecture is sticky — a cloud transcription business is built around that server, its billing, and its data pipeline, so it keeps the work in the cloud even when the device in your hands could do it. The practical result is that the moment you’re offline, or on a network that blocks the upload, the tool is dead weight.

There’s a second cost that has nothing to do with connectivity: everything you transcribe passes through someone else’s infrastructure. For a casual voice note that’s a shrug. For a client call, a patient conversation, or a confidential interview, it’s a real exposure — covered in depth in on-device meeting transcription. Offline transcription solves both problems with the same move: keep the work on the device.

What “offline” actually requires

Offline transcription has a strict definition, and it’s worth being precise because plenty of apps describe themselves as “private” or “secure” while still phoning home. For transcription to be genuinely offline, three things have to be true:

The speech model runs locally. The audio-to-text conversion happens on your Mac’s processor and Neural Engine, not on a remote GPU.
No audio leaves the device. Not for transcription, not for “quality improvement,” not for backup. The recording stays on your disk.
It works in airplane mode. The real test. Turn off wifi, turn off cellular if it’s a laptop with a modem, and the app still produces a transcript. If it can’t, it was never offline.

A tool that meets all three is doing the work on your hardware. A tool that fails the airplane-mode test is a cloud service with a local-looking front end, no matter what the marketing says.

How a Mac transcribes offline now

The thing that changed is that Apple put a production-grade speech model into the operating system. On macOS 26 Tahoe, the SpeechAnalyzer framework transcribes audio entirely on-device, running on the Neural Engine of any Apple-silicon Mac. There’s no API call, no upload, no account check. It’s a local model the same way the keyboard’s autocorrect is local.

Two things make this more than a checkbox feature:

It’s fast. On Apple silicon, SpeechAnalyzer runs faster than real time — a one-hour recording transcribes in a fraction of that — and per Apple’s published benchmarks it’s roughly 55% faster than Whisper v3 Turbo on the same chip. Offline transcription used to mean “accurate but slow.” On a modern Mac it’s accurate and fast.

The summary is local too. Transcription is only half the job; usually you also want a summary and the key points, not a wall of raw text. Apple’s on-device Foundation Models handle that part locally as well. So the entire pipeline — audio in, transcript and structured notes out — runs without a network round trip. Nothing about the recording or its text touches a server, on a plane or anywhere else.

The hardware requirement is the catch worth stating plainly: this needs an Apple-silicon Mac (M-series) on macOS 26. Apple Intelligence and SpeechAnalyzer’s on-device path don’t run on Intel Macs. If you’re on Apple silicon and current macOS, you already have the engine; you just need an app that uses it instead of a cloud backend.

Doing it with Dictanta

Dictanta is built on exactly this local pipeline, so offline is the default rather than a fallback. The flow for transcribing existing audio offline is short:

Import or record the audio. Drop in an .m4a, .wav, or .mp3 file, import an Apple Voice Memo, or record directly. None of this needs a connection.
Transcribe on-device. SpeechAnalyzer produces the transcript locally on the Neural Engine. Watch it run in airplane mode if you want proof — the text appears with the wifi off.
Get structured notes. On-device Foundation Models generate a summary and pull out the key points, so you end up with notes, not just a transcript dump.
Export. Save as Markdown, plain text, or a subtitle file. All local, all offline.

Because there’s no server in the loop, there’s also no upload queue and no per-minute cloud cost. The only limit on how much you can transcribe in a session is how fast the chip chews through the audio, which on Apple silicon is faster than real time. You can clear a stack of recordings on a flight and land with all of them as text.

The situations where offline is the whole point

Offline transcription isn’t a niche preference — it’s the difference between “can do the work” and “can’t” in a surprising number of real situations:

On a plane. The classic case. Airplane wifi is expensive, slow, and often blocks large uploads anyway. An offline transcriber turns a flight into the time you finally process those recorded interviews. The audio and the transcript both stay on the laptop on your tray table.

In the field. Field researchers, journalists, and inspectors record in places with no reliable signal — a remote site, a rural interview, a building with no reception. Offline means you transcribe where you are, not “once you get back to the hotel wifi.”

In secure environments. Some workplaces block cloud uploads at the network level, or policy forbids putting recordings through a third-party service. A clinic, a law firm, a government office. On-device transcription sidesteps the whole question: there’s no upload to block, because the audio never tries to leave. This is the same reason private transcription matters for meetings.

On bad wifi. You don’t have to be fully offline for cloud transcription to be miserable. Flaky hotel wifi, a packed conference network, a coffee shop at lunch — any of these can make an upload-based tool stall or fail mid-file. Local transcription doesn’t care; it never touches the network.

Offline vs. “private” vs. “secure” — read the fine print

Because “private” and “offline” sound interchangeable in marketing, it’s worth knowing what to check. An app can be encrypted in transit and at rest and still send your audio to a server — that’s “secure,” not offline. An app can promise it deletes your data after 30 days and still have had your recording on its infrastructure for those 30 days — that’s a retention policy, not on-device.

The only claim that means your audio never leaves the machine is the offline one, and the only way to verify it is the airplane-mode test. Turn off the network and try to transcribe. A genuinely offline tool keeps working. Everything else stops. It’s the cleanest test there is, and it costs you nothing to run before you trust a tool with anything sensitive.

What offline transcription is good at — and its limits

Offline on-device transcription is excellent at:

Pre-recorded audio. Interviews, voice memos, lectures, dictation. You have the file; it becomes text locally. This overlaps with transcribing Voice Memos to Markdown, which is the same offline pipeline pointed at your Voice Memos library.
Single-speaker and clean two-person audio. Dictation and clear interviews transcribe cleanly.
Anything you can’t or won’t upload. The confidential, the connectionless, the policy-restricted.

The honest limits:

Source audio quality still matters. Offline doesn’t mean magic. A recording made in a noisy room or far from the mic transcribes worse than a clean one, the same as any transcription. Dictanta’s editor lets you fix errors before exporting, but better input means less editing.
It needs the hardware. Apple-silicon Mac on macOS 26. No Intel Macs.
Speaker labels aren’t here yet. If you need a transcript that tags who said which line, that’s diarization, and it isn’t shipped in Dictanta today — it’s planned for a later version. For now the transcript is the words, accurately and locally, without per-speaker labels.

Where meeting capture fits

One distinction trips people up. Transcribing a file offline and recording a live meeting are related but not identical. If your goal is to capture a Zoom or Teams call — both your voice and the far-end audio coming out of your speakers — that’s system-audio capture, a Mac-specific capability covered in record system audio on a Mac and the no-bot Zoom guide. The transcription that follows the recording is the same offline pipeline described here; the capture step is what differs. If you already have the audio, you’re in the offline-transcription workflow this post is about. If you’re about to be on a call, you want meeting recording first, then the offline transcription happens automatically afterward.

Bottom line

Offline transcription on a Mac stopped being a compromise the moment Apple shipped SpeechAnalyzer and on-device Foundation Models in macOS 26. On an Apple-silicon Mac you can now turn speech into structured text faster than real time, with no internet and no audio ever leaving the machine — which means it works on a plane, in the field, on bad wifi, and inside networks that block the cloud entirely. The airplane-mode test is the proof: a real offline tool keeps transcribing with the network off.

If you want that without stitching frameworks together yourself, Dictanta does the whole offline pipeline — import or record, transcribe locally, summarize on-device, export — and it’s free for your first three recordings with no length cap, which is enough to run a real interview through it in airplane mode and watch the text appear with the wifi off. Paid tiers are $9.99/mo, $79.99/yr, or $149.99 lifetime.